645P The utility and limitations of artificial intelligence in diagnosing rare disorders

• Published in: Neuromuscular disorders : NMD 2024-10, Vol.43, p.104441, Article 104441.171
• Main Authors: Alawneh, I., Gonorazky, H., Repalo, A., Mashouri, P.
• Format: Article
• Language: English
• ISSN: 0960-8966
• DOI: 10.1016/j.nmd.2024.07.180

Genetic neuromuscular diseases (gNMD) represent a highly diverse spectrum of disorders encompassing over six hundred genes. Despite remarkable advancements in diagnostic modalities, a substantial proportion of gNMD patients, ranging from 20% to 60%, remain without a molecular diagnosis. This challenge is compounded by the presence of nonspecific clinical manifestations and overlapping phenotypes in some cases. Muscle magnetic resonance imaging (MRI) has emerged as a valuable diagnostic tool for assessing muscle involvement and severity patterns. However, its utilization has been limited to a select few expert centers globally due to the requisite expertise and time-intensive analysis process. In pursuit of a solution, we embarked on a journey harnessing the power of artificial intelligence and machine learning (AI) for image recognition to enhance the interpretation and comprehension of muscle imaging in hereditary myopathies. As AI is directly depending on the amount of data supplied a limitation of the application of AI in gNMD is the amount of data itself since all these disorders are rare. Leveraging publicly accessible database, we applied cutting-edge AI algorithms to refine the accuracy of the predictive results. We identified the limitations in using AI on muscle MRI including, time-consuming, missing data, non-homogeneous disease stage among patients, observer bias. Thereafter, possible real solutions were provided for each limitation with an ultimate goal to create the ideal database geared with innovative novel AI and machine learning techniques for automatic muscle MRI segmentation, automatic prediction, bypassing observer bias and save efforts and time.